![]() |
Alaska "Inside Passage" Copyright Andy Richards 2010 - All Rights Reserved |
![]() |
Lone Pine Farm - Kirby, Vermont Copyright Andy Richards 2023 - All Rights Reserved |
![]() |
Old Barn and Fall Foliage - Kirby, Vermont Copyright Andy Richards 2023 - All Rights Reserved |
I find that kind of annoyingTHE PERCEPTION that AI is "new" is not really an accurate one. It seems like the term is mentioned everywhere these days and every time someone doesn't like something about an image they see on line. I began to really see the explosion of commentary around late 2023 - 2024. But Adobe Photoshop has been introducing some iteration of "AI-based" tools for years, dating back at least to 2010 when their "content-aware" technology came out. Today, virtually (see what I did there? 😁) every software program now touts its "AI" features - often as if it were something brand new they they came up with. I find that kind of annoying.
![]() |
Florida Gulf Coast Sunset Copyright Andy Richards 2024 - All Rights Reserved |
THE DIFFERENCE today is that (perhaps due as much as anything to social media), people have become aware that it is a thing. For years, critics called imaged that were processed digitally "manipulated." As if the word is a bad word (as is certainly can be). The tried and true artist response to those criticisms has been to point out what film artists like Ansel Adams did to "manipulate" images in the darkroom. Today, though, we have computer-learning technology which applies accumulated examples inputed by humans to a data base that has "learned" how to apply those examples contextually. Is that really "intelligence" or is it just continually more focused iteration? I don't know the answer. What I do know is that however you want to characterize it, it just keeps improving its application.
![]() |
Swiss Alps - Lungern, Switzerland Copyright Andy Richards 2024 - All Rights Reserved |
![]() |
Pudding Hill Road in Vermont's "Northeast Kingdom" Copyright Andy Richards 2024 - All Rights Reserved |
IT ALSO is a ways from being perfected. The Golden Retriever photo here is an example of how it is not fully integrated between the data base and suggested input. In that image, I used my own background image of a back road in Vermont in fall, and queried Photoshop's AI - based "generative image" tool (found under the Edit dropdown menu). I tried a number of different phrases, and PS tried a few different times to generate an acceptable image. It never got it right. During the process, the image here is the one I found the most amusing example.
I have consistently chosen not to "render in the cloud"
CLOSELY RELATED to the "generate image" tool is a new component of the "crop" tool. Previously, we had standard cropping and another AI-based dropdown version that used "content aware" technology (again a circa 2010 addition to Photoshop). Much more recently, we now have a third choice: "generative expand," which works by adding canvas and then using the exact same interface as the "generate image" tool. Rather than supplying query terms, however, you just leave that box empty and instruct PS to generate and it will fill in the canvas contextually. Very much like content aware. But more sophisticated. It does a surprisingly good job. Each time you ask it to generate, it gives three choices. If you don't like any of them, tell it to do it again (though in my own brief empirical testing the first one is usually the one I like best). There is one negative (which I suspect will be overcome in the not too distant future). The generated image area is not very high resolution. I have used "generative expand" on a few occasions when I didn't frame an image as well as I would have liked. You can notice the difference in resolution. This would probably be an issue for large prints - though there are some workarounds.
ONE OTHER concern comes to mind here. Most of the new software offering some kind of "AI" capability offers two different "spaces" for conducting the rendering: on your computer and in "the cloud" (which by now everybody knows is euphemistic terminology for "their computer"). As I mentioned earlier, the current delivery method of "AI" generation involves an image memory bank of thousands of images, with an algorithym directing the computer to compare some combination of your image and/or your verbal instructions to the image bank, searching for relevant similar imagery and creating the AI rendering. That is about as elementary (and probably inaccurate in many ways) an explanation as I can give (and probalby the best I am capable of giving). But it does set up my explanation of the two choices. If you choose to render your AI in "the cloud," you will have many more and probably more current imagery to draw from, as well as the quite probable greater computing power of the providers' computers. If you keep it "in house" your comparison bank will be (somewhat) smaller.
![]() |
Mala Strana; Prague, Czech Republic Copyright Andy Richards 2024 - All Rights Reserved |
![]() |
Weston, Vermont Copyright Andy Richards 2010 - All Rights Reserved |
I HAVE consistently chosen not to "render in the cloud." I think there are some potential negatives. First, many providers are now charging for each render in some fashion (even subscription based programs will be issuing a set amount of "credits" against cloud-based "AI" rendering and when you run out, you will have to purchase more. Second, in this day and age, I have a healthy fear of just putting my work out there on the cloud. I am certain that among other things, the cloud based AI bank is augmenting each time you put an image out there. I might be wrong. 😏
![]() |
Spit Rock Light - Split Rock State Park - Two Harbors, Minnesota Copyright Andy Richards 2010 - All Rights Reserved |
EITHER WAY, "AI" is here to stay, and will continue to grow its presence in our lives, and in particular in the way we post-process our images. In future blogs, I will post a couple additional blogs that will cover other specific AI tools that are offered by Photoshop, and how well they work (or don't). Every one of the photographs in this post have some element of "AI" purposely implemented in them Can you identify what was done to each?
No comments:
Post a Comment
PLEASE COMMENT